Both trace agents and score production traffic, but they start from opposite ends of the release. LangSmith is a development workbench with datasets, experiments, your own judges and prompt versioning; Raindrop watches live traffic for behaviours you describe and groups failures into issues. Where your agent is in its life usually decides it.[1][4]
LangSmith is LangChain's platform for building and testing agents, with datasets, experiments, evaluators you write yourself and prompt versioning; Raindrop is a hosted monitor for agents already in production that classifies live events with custom signals and groups recurring failures into issues. LangSmith prices per seat plus traces, Raindrop per event. Both companies are independent, and neither can stop a tool call during a run.
| Decision factor | LangSmith | Raindrop |
|---|---|---|
| Ownership | LangChain, independent; $125M Series B in October 2025 | Independent; Series A led by CRV in September 2026, $50M raised in total |
| Centre of gravity | Datasets, experiments and evaluators across the build loop | Production events: Signals, Issue Detection and Experiments on live traffic |
| Evaluators | LLM-as-judge, pairwise and code evaluators, offline and online | Signals drafted from plain language by Raindrop's own models; no custom judge prompt or model |
| Prompt tooling | Prompt Hub with commit-based versioning and a playground | None; Experiments can compare prompt versions on production traffic |
| Self-hosting | Enterprise add-on, or a BYOC data plane | VPC deployment in beta for Enterprise and select partners |
| Run-time controls | Partial: LLM Gateway (beta) caps spend, restricts models and redacts model calls | None documented; analyses events after they arrive |
| Free tier | Developer: 1 seat, 5k base traces a month, 14-day retention | Hobby: 1,000 events a month, 1 custom signal, 14-day retention |
| First paid tier | Plus: $39 per seat a month, 10k base traces | Pro: $299 a month plus $0.003 per event up to 1M events |
| Capability | LangSmith | Raindrop | Prefactor |
|---|---|---|---|
| Observe | |||
| LLM call tracing | ✓ | ✓ | ✓ |
| Multi-step agent tracing (tool calls, nested spans) | ✓ | ✓ | ✓ |
| Cost tracking | ✓ | ✓ | ✓ |
| OpenTelemetry support | ✓ | ✓ | ✓ |
| Evaluate | |||
| Offline evals (datasets, experiments) | ✓ | Partial | ✓ |
| Online evals on production traffic | ✓ | ✓ | ✓ |
| LLM-as-a-judge scoring | ✓ | Partial | ✓ |
| Code-based evaluators | ✓ | Partial | ✓ |
| Human review and annotation | ✓ | Partial | ✓ |
| Prompt versioning and playground | ✓ | — | — |
| Act on findings | |||
| Runtime controls on live agent actions | Partial | — | ✓ |
| Gateway or proxy in the request path | — | — | — |
| Licence and deployment | |||
| Open source | Partial | Partial | Partial |
| Self-hosting | ✓ | Partial | Partial |
LangSmith covers the loop a team runs while building an agent: trace a run, add it to a dataset, run an experiment against a new prompt or model, compare the results, then score live traffic with the same evaluators. Online evaluators run through automation rules with filters and sampling rates, including on multi-turn threads, and production traces can be pulled back into datasets.
Raindrop is built for the period after launch. Its Signals are classifiers that run on every new event and can backfill 90 days of sampled history. Issue Detection groups recurring failures, and Experiments compare signal rates between cohorts of production traffic, so a change is judged on real users rather than a fixed dataset. Raindrop has no hosted offline eval runner. Its open-source Workshop runs on a developer's machine and has a coding agent turn failures into evals, and Simulations, which test a change on each pull request, are in early access.
If you are pre-launch and still shaping prompts, LangSmith fits that work. If an agent is already serving users and the open question is what keeps going wrong, Raindrop is aimed at exactly that.
LangSmith lets you write the evaluator yourself: an LLM-as-judge prompt, a pairwise comparison or a code check, used offline and online. Annotation queues support single-run and pairwise review with rubrics, required reviewer counts and assigned reviewers. LangSmith Engine, in public beta since May 2026, reads production traces, opens issues and drafts fixes; the September 2026 version added red teaming and automated fix testing.
In Raindrop you describe a behaviour in plain language, label sample events as Match or Not Match, and Raindrop builds the classifier. Since August 2026 that runs on rd-signal-2, which combines generated filter code, a task-specific classifier and an in-house reasoning model for ambiguous cases. You cannot bring your own judge prompt or model. That means less set-up work and less say over how a score is reached. Human review is lighter as well: dataset members can be labelled Good, Bad or Needs review, but there is no dedicated reviewer queue.
As of 20 October 2026, LangSmith's Developer tier is free for one seat and 5k base traces a month, and Plus is $39 per seat a month with 10k base traces. LangSmith does not publish its per-trace overage rate, and products such as Engine and Deployment are metered in LangChain Standard Units. Self-serve customers moved to new plans this month.
Raindrop charges by event. Hobby is free for 1,000 events a month, after which ingestion stops until the next month, with one custom signal and 14-day retention. Pro is $299 a month plus $0.003 per event up to 1M events, then $0.002, with ten custom signals, unlimited retention and a 14-day trial. PII redaction is a separate add-on from $80 a month. A bigger team costs more on LangSmith; more traffic costs more on Raindrop. Check LangSmith's pricing and Raindrop's plans against your own seat count and event volume.
Both platforms are proprietary and hosted by default. Self-hosted LangSmith is an Enterprise add-on deployed on Kubernetes, and a BYOC data plane is also Enterprise only. Raindrop's VPC deployment is in beta and rolling out to Enterprise customers and select partners. Workshop is MIT-licensed, but it is a local debugger, not a copy of the platform.
On stack fit, LangSmith has the deepest integration with LangChain, LangGraph and Deep Agents, with SDKs for Python, TypeScript, Go and Java. Raindrop lists a long set of framework integrations, including LangChain, Deep Agents and the Claude Agent SDK, though most are marked beta. Both accept OpenTelemetry traces; Raindrop takes OTLP over HTTP only, with no gRPC, metrics or logs.
Neither stops an agent's action during a run. LangSmith's LLM Gateway, in public beta since 30 July 2026, can cap spend, restrict models and redact personal data in model calls, but tool and MCP gateways are described as planned. Raindrop analyses events after ingestion and sends alerts.
We sell the layer this section describes. Read it with that in mind.
Stated plainly, since Prefactor publishes this page: Prefactor is an agent evaluation platform that scores each step of a run and can act on what it finds through runtime policies on tool calls, approval holds for risky actions and a kill switch for a run, an agent or a team. Raindrop documents no run-time controls, and LangSmith's beta gateway covers model calls but not tool calls. Prefactor has no prompt playground or prompt registry and its platform is hosted, so if prompt iteration is the daily work, LangSmith is the better fit, and if you want failures clustered from plain-language descriptions, Raindrop is. See the full comparison of 15 platforms.
Reviewed against public product and documentation pages on October 20, 2026. If a vendor has changed a feature, product name, or positioning since then, send a correction and we will update it. Numbered source links in the page body point to the ordered sources below.
Prefactor scores each step of every agent run and can hold or stop an action when a score says it should not happen. Start on the free tier: 5,000 spans a month, no credit card required.
Prefactor helps teams observe, evaluate, and improve their AI agents in production, across every framework and provider.